Hyperspectral application for locust monitoring: algorithm development and verification

Locusts, including migratory (Locusta migratoria) and desert locusts (Schistocerca gregaria), occasionally form large swarms and experience outbreaks. Large-scale locust outbreaks pose serious threats to agricultural crops. Early detection and termination are critical for mitigating damage. Currently, locusts are primarily monitored through ground surveys, which makes continuous observation over large areas challenging. Therefore, the development of remote-sensing technologies for estimating locust distribution is highly desirable. Hyperspectral (HS) remote sensing offers the potential to detect and identify ground targets such as locusts. Previous studies attempted to estimate the locust coverage ratio (LCR) using HS data through optimization algorithms based on numerical methods. A numerical optimization-based algorithm was developed previously to detect locusts and estimate their LCR using an HS sensor. Although the previous implementation executed the optimization numerically, in this study, an analytical solution to the optimization problem was derived to refine the algorithm. The proposed algorithm estimates the LCR with higher accuracy and greater computational efficiency than the previous method and eliminates pixel regions where the LCR is, by design, undefined under that method’s specification. We refer to this algorithm as the spectral optimization-based target abundance estimator (SOTAE). The SOTAE was verified through simulation. SOTAE can detect locusts with an area under curve of 0.9 when the ground LCR exceeds 0.02, a 2.5-fold lower LCR than in the previous study. Moreover, the computation duration was 0.005 times shorter than that of the previous method. Overall, SOTAE is a fast and accurate algorithm that shows potential for locust monitoring.